No abstract

No abstract

Power grid (PG) simulation is critical for verification of supply noises in IC design. Computational demands for simulating PG is high. Cloud computing can be leveraged to mitigate these costs. However, simulating on third-party platforms lead to major security concerns. We propose a framework for secure PG simulation on Cloud. Multiple compression strategies are employed to reduce communication overhead. Turnaround time similar to an insecure simulator on Cloud can be achieved, while securing current excitations and output voltage vectors at reasonable costs.A typical PG is modeled as an RC/RLC circuit with current sources [1,2], and formulated as a system of differential algebraic equations (DAE). With increasing verification complexity, users are forced to spend heavily on computing resources. Cloud computing offers a scalable, customizable and inexpensive computing platform that can be rapidly provisioned and released [3]. Cloud computing for PG simulations raises security concerns, since simulation data typically carry sensitive design information.We investigate secure outsourcing of system of DAE to a Cloud. Excitations can be transformed, such that all computations are performed on the transformed patterns. Cloud has no access to inputs and outputs of the simulator. This secures the temporal and spatial power consumption profile of the IC. It can be extended to simulate the PG under multiple excitations simultaneously. Our compression algorithms can reduce communication costs by compressing data by 95-99% without affecting the performance. Our approach can be extended to hide the coefficients of equations, thus securing circuit parameters associated with the PG network.The paper is organized as follows. Problem statement and design goals are introduced in Section 2. Our proposed solution is presented in Section 3. Section 4 depicts our experimental results, followed by our concluding remarks in Section 5. PROBLEM STATEMENT & DESIGN GOALS 2.1. Power Grid ModelThe discrete system equation for an RC model with N non-Vdd nodes, using Backward Euler approximation, can be expressed as(1) where G is an NxN conductance matrix, C is an NxN diagonal matrix of capacitances, V n is an Nx1 vector of voltage drops and I n is an Nx1 vector of input current excitations at n th instant and h is the time step. Our transformation is also applicable for RLC models of PG, utilizing Trapezoidal Rule for discretization. Direct methods based on LU factorization or Cholesky decomposition for for solving the equations can be prohibitively resource hungry.Cloud outsourcing architecture involves two entities, illustrated in Fig. 1, user, and Cloud server (CS). CS refers to an Amazon EC2 instance [4]; a virtual server with predefined configuration. Fig. 1: PG Simulation OutsourcingUser provides input excitation vectors, in excitation files, at discrete instants to the simulator, running on an EC2 instance. Voltage noises computed by the simulator are written to output files. The overall approach can be divided into three phases: curre...

Efficient power grid verification, critical in modern integrated circuits, is computationally demanding because of increasing grid sizes. Vectorless approach to grid verification estimates worst-case voltage noises without detailed evaluation of load current excitations. We study voltage noise assessment in RLC models of VDD and GND networks in integrated power grids. Transitory circuit behaviours are captured by transient constraints, while abstract grid model is utilized to accelerate convergence. Heuristics are proposed to extract constraints that restrict power consumption profiles to realistic scenarios. Bounds on voltage overshoots and undershoots are evaluated by formulating multiple optimization problems. We propose ways to mitigate storage and computational requirements on processing resources, enabling users to deploy computations on economical Cloud Computing platforms. Recommended solution is parallelizable, thereby reducing the overall verification time. Experimental results suggest that the proposed technique is practical and scalable for industrial grids.